Zina Ibrahim - Big Data in Mental Health - 23rd July 2014

Post on 07-May-2015

456 views 1 download

description

Organised by the Bioinformatics group at the BRCMH, IoP, SLaM and Maudsley Digital, this symposium showcased talks regarding the important roles of big data in mental health biomedical research and treatments.

Transcript of Zina Ibrahim - Big Data in Mental Health - 23rd July 2014

Multi-agent Systems for Automating Large Naturalistic Treatment Trials in Routine Practice

Zina Ibrahim

Motivating Research❏ By: Lorena Fernández de la Cruz, Argyris Stringaris, Robert Goodman and

others at the Department of Child and Adolescent Psychiatry, Institute of Psychiatry, King’s College London.

❏ Aim: Conduct large naturalistic treatment trials on ADHD ❏ Test the effectiveness of ADHD treatments in routine clinical settings❏ Establish the effectiveness of ADHD treatments for the subgroup of

children with mood dysregulation.

❏ Assessment Variable: Strength and Difficulties Questionnaire clinical outcome measure (SDQ) scores in the de-identified electronic health records.

The Strengths and Difficulties Questionnaire (SDQ)

❏ A brief child mental health questionnaire for children and adolescents ages 2 through 17 years old, developed by Dr. Robert Goodman.

❏ Measures 25 attributes, some positive and others negatives.❏ The 25 attributes fall into the following 4 scales:

❏ Emotional Symptoms (5 items)❏ Conduct problems (5 items)❏ Hyperactivity/inattention (5 items)❏ Peer relationship problems (5 items)❏ Prosocial behaviour (5 items)

❏ The patient is given a score for each attribute. The total of all the scores provides the SDQ score of the patient.

Outcome of Initial Research

❏ Excellent baseline SDQ

data available

❏ Baseline SDQ scores

predictive of diagnosis

outcome

Number of EPJS Records for ADHD Cases 8,434

Gender (% Female) 44%

Age: M (SD) 11.2 (3.8)

Records with SDQ 6, 912

Problem: No Treatment Outcome Measures

❏ Apart from the initial diagnosis SDQ, few patients have repeated SDQ measures, collected at subsequent intervals, to assess treatment outcomes.

Subsample of ADHD individuals with at least 2 SDQ

N:157 Age: M(13) Gender: 82% F

Medication: 148 valid, 9 missing

Medication N %

Methylphenidate 137 87.3

Dextroamphetamine 1 0.6

Atomoxetine 29 18.5

Clonidine 9 5.7

Number of Meds N %

1 121 77.1

2 26 16.6

3 1 0.6

Solution: Automatic Collection of Outcome Measures

❏ Aims: ❏ Improve the quality of treatment outcome measure (SDQ)

collection in EPJS (and subsequently CRIS). ❏ Engage with patients through online web resources.

❏ Objectives: ❏ Automate clinical trial feedback without clinicians’ intervention.❏ Provide patients with detailed analyses based on their SDQ

scores. The analyses provide books, helpful hints and link to help them better understand their situations.

❏ Description: build an automated computer system which: ❏ Regularly queries EPJS for newly-created first (baseline) SDQ entries❏ Creates virtual agents (autonomous software components) for every

SDQ entry found. Every agent will: ❏ Generate a personalised guidance report using youthinmind.info based on

the SDQ entries❏ Make the following available on an online web resource:

❏ The personalised guidance report❏ Forms for filling new SDQ entries

Solution: Automatic Collection of Outcome Measures

❏ Create a monthly follow-up schedule for the case. Every month the virtual agent will:

❏ Monitor adherence to the follow-up schedule by regularly sending reminders to participants to complete the web form until one is filled

❏ Once filled, the agent sends the new SDQ entries for the case from the web resource to EPJS

❏ Generate a new personalised guidance report based on the new entries (from youthinmind.info)

Solution: Automatic Collection of Outcome Measures

❏ A Software Agent: is a computer program which:❏ Acts autonomously on behalf of its user❏ Acts proactively to achieve a predefined goal❏ Reacts to input from the changing environment❏ Is social, i.e. it communicates with other agents

❏ A Multi-agent system: computer system made of a number of software agents jointly interacting to achieve the design requirements of the overall system.

Multi-Agent Systems

Multi-Agent Systems for Generating Treatment Trials

Overall System Architecture

A Month after last SDQ is entered

Weekly reminders until user signs up and fills the next SDQ form.

Repeat for six SDQs

youthinmind.info

Implementation and Progress

❏ Implementation Details: ❏ JADE (Java-based Agent Development Environment) Java-based environment

which interacts with the user through Servlets and JSP pages. ❏ Prototype developed based on the EPJS Testing database

❏ Development Progress:❏ Core functionalities 95% complete❏ Currently working on:

❏ Parsing patient/informer e-mail addresses from free-text data❏ Interacting with the users using text messaging for reminders and alerts. ❏ Designing a (good looking) web interface.

❏ Post-development Phase❏ Acquisition of permissions for deployment over EPJS.

Acknowledgmentshttp://core.brc.iop.kcl.ac.uk

Dr Richard J Dobson (BRC)Dr Lorena Fernandez de la Cruz (IoP)Dr Argyris Stringaris(IoP)Prof. Robert Goodman (IoP)Prof. Emily Siminoff (IoP)Prof. Andrew Pickles (IoP)Dr Matthew Broadbent (BRC)Dr Caroline Johnston (BRC)Dr Amos Folarin (BRC)John Turp (SLAM)